Evidence before assertion
Claims should be connected to data, methods and clearly stated limits.
Applied AI research · Istanbul
Advancing intelligence for financial systems.
Colendi AI is an applied research laboratory developing models, agents, evaluation methods and intelligent systems for the future of finance.
Constraint-aware reasoning / system map
RA–01
decision state reasoning trace control checkpoint
Financial foundation models
Domain intelligence for financial language, structured data and institution-specific knowledge.
02Agentic banking
Agents that can assist, analyse and act within explicit financial and operational boundaries.
03Credit intelligence
Decision systems that connect predictive performance with explainability, fairness and portfolio outcomes.
Colendi AI · research scope
Research agenda · 01
A financial answer is only useful when its decision path can be examined.
Constraint-aware reasoning / system map
RA–01
decision state reasoning trace control checkpoint
In preparation
Financial Reasoning Under Regulatory Constraints
Accuracy alone is an incomplete measure for financial AI. Our research agenda separates evidence retrieval, policy adherence, calculation, uncertainty and accountable action.
- Focus
- Traceable financial reasoning
- Methods
- Scenario evaluation · decision traces
- Languages
- Turkish · English
- Stage
- Protocol and dataset design
Research principles
Research must survive contact with the real world.
Evaluation before deployment
A model is not ready because it can answer; it is ready when its failure modes are understood.
Humans remain accountable
Automation should make responsibility more legible, never hide it inside a system.
Systems, not demos
We study the complete decision path: model, data, policy, interface, people and operations.
P–01
See how the technology is delivered Credit intelligence
Decision support for more contextual, reviewable credit assessment.
P–02Behavioral segmentation and dynamic limits
Customer understanding that can respond to changing financial behaviour.
P–03Fraud detection and risk monitoring
Signals that help institutions identify anomalous activity and changing risk.
P–04AI assistants and controlled financial agents
Assistants that support people and processes within explicit boundaries.
E–01
Intelligence layer
Colendi AI
Research & systemsE–02Regulated banking layer
ColendiBank
Live institutionE–03Embedded finance layer
Colendi Platform
Live platformE–04Capital-markets layer
Colendi Menkul
Live institutionE–05Insurance-distribution layer
Colendi Sigorta / COMPI
Live productGroup scale · publicly stated by Colendi Group
18M+users across the group's marketsColendi Group
3,500data points in the group scoring systemColendi Group
5connected financial layersColendi ecosystem
S–01
Financial intelligence platform
ColendiMind
The applied platform connecting language models, institution-specific knowledge and financial workflows.
Deployed systemS–02Decision science
Credit Intelligence
Models and monitoring methods for behavioural credit assessment, limit decisions and portfolio learning.
Applied researchS–03Agent systems
Financial Agents
Controlled assistants and agents for analysis, service and internal financial operations.
Research prototypesS–04Research infrastructure
Evaluation Lab
Test harnesses for reliability, policy adherence, multilingual performance and decision traceability.
In developmentC–01Live system
Agentic insurance · COMPI · Colendi Sigorta
A voice agent at the front door of insurance
Turn a static lead form into an immediate, consent-led conversation.
COMPI invites a customer to leave their details, captures explicit consent and initiates a voice-AI call designed to understand needs and answer product questions.
Open COMPI caseLIVE PRODUCT JOURNEYCOMPI / C–01
01ConsentIdentity + permission
02ConnectOutbound AI call
03UnderstandNeeds + questions
04ActQuote + handoff
Research agenda · 01
In preparationFinancial Reasoning Under Regulatory Constraints
A research programme for measuring whether financial AI systems can connect an answer to evidence, policy and an accountable decision path.
Colendi AI Research2026
Methods note · 02
In preparationBeyond Answer Accuracy: Evaluating the Decision Trace
A proposed evaluation structure for separating evidence retrieval, policy adherence, calculation and final recommendation quality.
Colendi AI Research2026
Applied research note · 03
In preparationHuman Approval as a System Component
An examination of where human judgement adds control, where it becomes ceremonial and how agent interfaces can preserve real accountability.
Colendi AI Research2026
Research with us
Finance gives AI some of its hardest—and most consequential—problems.
We welcome research conversations with scientists, engineers, financial institutions and public-interest organisations.
Start a conversation